Automatic building a corpus and exploiting it for polarity classification of indirect opinions about drugs

Opinion mining is a well-known problem in natural language processing that has attracted increasing attention in recent years. Existing approaches have been often focused on identifying direct opinions and ignored indirect ones. However, in some domains such as medical, indirect opinions occur frequently. Therefore, ignoring indirect opinions can lead to the loss of valuable information and noticeable decline in overall accuracy of opinion mining systems. In this paper, we present a semi-automatic approach to construct a corpus of indirect opinions from drug reviews. In the first step, we propose an automatic method for detection of indirect opinions and in the second step, we use domain knowledge, linguistic rules and review structure for polarity detection of drug reviews. Then we exploit the constructed corpus as a training set in machine learning techniques for polarity classification of new examples. Experimental results demonstrate that our proposed approach achieves 82.81 percent precision.